In many applications, a dataset can be considered as a set of observed signals that live on an unknown underlying graph structure. Some of these signals may be seen as white noise that has been filtered on the graph topology by a graph filter. Hence, the knowledge of the filter and the graph provides valuable information about the underlying data generation process and the complex interactions that arise in the dataset. We hence introduce a novel graph signal processing framework for jointly learning the graph and its generating filter from signal observations. We cast a new optimisation problem that minimises the Wasserstein distance between the distribution of the signal observations and the filtered signal distribution model. Our proposed method outperforms state-of-the-art graph learning frameworks on synthetic data. We then apply our method to a temperature anomaly dataset, and further show how this framework can be used to infer missing values if only very little information is available.
Modeling patient disease progression using Electronic Health Records (EHRs) is critical to assist clinical decision making. While most of prior work has mainly focused on developing effective disease progression models using EHRs collected from an individual medical system, relatively little work has investigated building robust yet generalizable diagnosis models across different systems. In this work, we propose a general domain adaptation (DA) framework that tackles two categories of discrepancies in EHRs collected from different medical systems: one is caused by heterogeneous patient populations (covariate shift) and the other is caused by variations in data collection procedures (systematic bias). Prior research in DA has mainly focused on addressing covariate shift but not systematic bias. In this work, we propose an adversarial domain separation framework that addresses both categories of discrepancies by maintaining one globally-shared invariant latent representation across all systems} through an adversarial learning process, while also allocating a domain-specific model for each system to extract local latent representations that cannot and should not be unified across systems. Moreover, our proposed framework is based on variational recurrent neural network (VRNN) because of its ability to capture complex temporal dependencies and handling missing values in time-series data. We evaluate our framework for early diagnosis of an extremely challenging condition, septic shock, using two real-world EHRs from distinct medical systems in the U.S. The results show that by separating globally-shared from domain-specific representations, our framework significantly improves septic shock early prediction performance in both EHRs and outperforms the current state-of-the-art DA models.
We investigate the problem of training neural networks from incomplete images without replacing missing values. For this purpose, we first represent an image as a graph, in which missing pixels are entirely ignored. The graph image representation is processed using a spatial graph convolutional network (SGCN) -- a type of graph convolutional networks, which is a proper generalization of classical CNNs operating on images. On one hand, our approach avoids the problem of missing data imputation while, on the other hand, there is a natural correspondence between CNNs and SGCN. Experiments confirm that our approach performs better than analogical CNNs with the imputation of missing values on typical classification and reconstruction tasks.
Continuous input signals like images and time series that are irregularly sampled or have missing values are challenging for existing deep learning methods. Coherently defined feature representations must depend on the values in unobserved regions of the input. Drawing from the work in probabilistic numerics, we propose Probabilistic Numeric Convolutional Neural Networks which represent features as Gaussian processes (GPs), providing a probabilistic description of discretization error. We then define a convolutional layer as the evolution of a PDE defined on this GP, followed by a nonlinearity. This approach also naturally admits steerable equivariant convolutions under e.g. the rotation group. In experiments we show that our approach yields a $3\times$ reduction of error from the previous state of the art on the SuperPixel-MNIST dataset and competitive performance on the medical time series dataset PhysioNet2012.
Time-series data with missing values are commonly encountered in many fields, such as healthcare, meteorology, and robotics. The imputation aims to fill the missing values with valid values. Most imputation methods trained the models implicitly because missing values have no ground truth. In this paper, we propose Random Drop Imputation with Self-training (RDIS), a novel training method for time-series data imputation models. In RDIS, we generate extra missing values by applying a random drop on the observed values in incomplete data. We can explicitly train the imputation models by filling in the randomly dropped values. In addition, we adopt self-training with pseudo values to exploit the original missing values. To improve the quality of pseudo values, we set the threshold and filter them by calculating the entropy. To verify the effectiveness of RDIS on the time series imputation, we test RDIS to various imputation models and achieve competitive results on two real-world datasets.
The decision tree is one of the most popular and classical machine learning models from the 1980s. However, in many practical applications, decision trees tend to generate decision paths with excessive depth. Long decision paths often cause overfitting problems, and make models difficult to interpret. With longer decision paths, inference is also more likely to fail when the data contain missing values. In this work, we propose a new tree model called Cascading Decision Trees to alleviate this problem. The key insight of Cascading Decision Trees is to separate the decision path and the explanation path. Our experiments show that on average, Cascading Decision Trees generate 63.38% shorter explanation paths, avoiding overfitting and thus achieve higher test accuracy. We also empirically demonstrate that Cascading Decision Trees have advantages in the robustness against missing values.
Drilling boreholes for gas and oil extraction is an expensive process and profitability strongly depends on characteristics of the subsurface. As profitability is a key success factor, companies in the industry utilise well logs to explore the subsurface beforehand. These well logs contain various characteristics of the rock around the borehole, which allow petrophysicists to determine the expected amount of contained hydrocarbon. However, these logs are often incomplete and, as a consequence, the subsequent analyses cannot exploit the full potential of the well logs. In this paper we demonstrate that Machine Learning can be applied to \emph{fill in the gaps} and estimate missing values. We investigate how the amount of training data influences the accuracy of prediction and how to best design regression models (Gradient Boosting and neural network) to obtain optimal results. We then explore the models' predictions both quantitatively, tracking the prediction error, and qualitatively, capturing the evolution of the measured and predicted values for a given property with depth. Combining the findings has enabled us to develop a predictive model that completes the well logs, increasing their quality and potential commercial value.
Real-world applications that involve missing values are often constrained by the cost to obtain data. Test-cost sensitive, or costly feature, methods additionally consider the cost of acquiring features. Such methods have been extensively studied in the problem of classification. In this paper, we study a related problem of test-cost sensitive methods to identify nearby points from a large set, given a new point with some unknown feature values. We present two models, one based on a tree and another based on Deep Reinforcement Learning. In our simulations, we show that the models outperform random agents on a set of five real-world data sets.
We consider the problem of estimating the conditional probability distribution of missing values given the observed ones. We propose an approach, which combines the flexibility of deep neural networks with the simplicity of Gaussian mixture models (GMMs). Given an incomplete data point, our neural network returns the parameters of Gaussian distribution (in the form of Factor Analyzers model) representing the corresponding conditional density. We experimentally verify that our model provides better log-likelihood than conditional GMM trained in a typical way. Moreover, imputation obtained by replacing missing values using the mean vector of our model looks visually plausible.
Cross-validation (CV) is a technique used to estimate generalization error for prediction models. For pipeline modeling algorithms (i.e. modeling procedures with multiple steps), it has been recommended the entire sequence of steps be carried out during each replicate of CV to mimic the application of the entire pipeline to an external testing set. While theoretically sound, following this recommendation can lead to high computational costs when a pipeline modeling algorithm includes computationally expensive operations, e.g. imputation of missing values. There is a general belief that unsupervised variable selection (i.e. ignoring the outcome) can be applied before conducting CV without incurring bias, but there is less consensus for unsupervised imputation of missing values. We empirically assessed whether conducting unsupervised imputation prior to CV would result in biased estimates of generalization error or result in poorly selected tuning parameters and thus degrade the external performance of downstream models. Results show that despite optimistic bias, the reduced variance of imputation before CV compared to imputation during each replicate of CV leads to a lower overall root mean squared error for estimation of the true external R-squared and the performance of models tuned using CV with imputation before versus during each replication is minimally different. In conclusion, unsupervised imputation before CV appears valid in certain settings and may be a helpful strategy that enables analysts to use more flexible imputation techniques without incurring high computational costs.
Functional Magnetic Resonance Imaging (fMRI) is a neuroimaging technique with pivotal importance due to its scientific and clinical applications. As with any widely used imaging modality, there is a need to ensure the quality of the same, with missing values being highly frequent due to the presence of artifacts or sub-optimal imaging resolutions. Our work focus on missing values imputation on multivariate signal data. To do so, a new imputation method is proposed consisting on two major steps: spatial-dependent signal imputation and time-dependent regularization of the imputed signal. A novel layer, to be used in deep learning architectures, is proposed in this work, bringing back the concept of chained equations for multiple imputation. Finally, a recurrent layer is applied to tune the signal, such that it captures its true patterns. Both operations yield an improved robustness against state-of-the-art alternatives.
Missing value imputation is crucial for real-world data science workflows. Imputation is harder in the online setting, as it requires the imputation method itself to be able to evolve over time. For practical applications, imputation algorithms should produce imputations that match the true data distribution, handle data of mixed types, including ordinal, boolean, and continuous variables, and scale to large datasets. In this work we develop a new online imputation algorithm for mixed data using the Gaussian copula. The online Gaussian copula model meets all the desiderata: its imputations match the data distribution even for mixed data, improve over its offline counterpart on the accuracy when the streaming data has a changing distribution, and on the speed (up to an order of magnitude) especially on large scale datasets. By fitting the copula model to online data, we also provide a new method to detect change points in the multivariate dependence structure with missing values. Experimental results on synthetic and real world data validate the performance of the proposed methods.
We propose a novel generalisation to the Student-t Probabilistic Principal Component methodology which: (1) accounts for an asymmetric distribution of the observation data; (2) is a framework for grouped and generalised multiple-degree-of-freedom structures, which provides a more flexible approach to modelling groups of marginal tail dependence in the observation data; and (3) separates the tail effect of the error terms and factors. The new feature extraction methods are derived in an incomplete data setting to efficiently handle the presence of missing values in the observation vector. We discuss various special cases of the algorithm being a result of simplified assumptions on the process generating the data. The applicability of the new framework is illustrated on a data set that consists of crypto currencies with the highest market capitalisation.
Multivariate time-series data are used in many classification and regression predictive tasks, and recurrent models have been widely used for such tasks. Most common recurrent models assume that time-series data elements are of equal length and the ordered observations are recorded at regular intervals. However, real-world time-series data have neither a similar length nor a same number of observations. They also have missing entries, which hinders the performance of predictive tasks. In this paper, we approach these issues by presenting a model for the combined task of imputing and predicting values for the irregularly observed and varying length time-series data with missing entries. Our proposed model (Bi-GAN) uses a bidirectional recurrent network in a generative adversarial setting. The generator is a bidirectional recurrent network that receives actual incomplete data and imputes the missing values. The discriminator attempts to discriminate between the actual and the imputed values in the output of the generator. Our model learns how to impute missing elements in-between (imputation) or outside of the input time steps (prediction), hence working as an effective any-time prediction tool for time-series data. Our method has three advantages to the state-of-the-art methods in the field: (a) single model can be used for both imputation and prediction tasks; (b) it can perform prediction task for time-series of varying length with missing data; (c) it does not require to know the observation and prediction time window during training which provides a flexible length of prediction window for both long-term and short-term predictions. We evaluate our model on two public datasets and on another large real-world electronic health records dataset to impute and predict body mass index (BMI) values in children and show its superior performance in both settings.
As the number of installed meters in buildings increases, there is a growing number of data time-series that could be used to develop data-driven models to support and optimize building operation. However, building data sets are often characterized by errors and missing values, which are considered, by the recent research, among the main limiting factors on the performance of the proposed models. Motivated by the need to address the problem of missing data in building operation, this work presents a data-driven approach to fill these gaps. In this study, three different autoencoder neural networks are trained to reconstruct missing short-term indoor environment data time-series in a data set collected in an office building in Aachen, Germany. This consisted of a four year-long monitoring campaign in and between the years 2014 and 2017, of 84 different rooms. The models are applicable for different time-series obtained from room automation, such as indoor air temperature, relative humidity and $CO_{2}$ data streams. The results prove that the proposed methods outperform classic numerical approaches and they result in reconstructing the corresponding variables with average RMSEs of 0.42 {\deg}C, 1.30 % and 78.41 ppm, respectively.
The problem of graph learning concerns the construction of an explicit topological structure revealing the relationship between nodes representing data entities, which plays an increasingly important role in the success of many graph-based representations and algorithms in the field of machine learning and graph signal processing. In this paper, we propose a novel graph learning framework that incorporates the node-side and observation-side information, and in particular the covariates that help to explain the dependency structures in graph signals. To this end, we consider graph signals as functions in the reproducing kernel Hilbert space associated with a Kronecker product kernel, and integrate functional learning with smoothness-promoting graph learning to learn a graph representing the relationship between nodes. The functional learning increases the robustness of graph learning against missing and incomplete information in the graph signals. In addition, we develop a novel graph-based regularisation method which, when combined with the Kronecker product kernel, enables our model to capture both the dependency explained by the graph and the dependency due to graph signals observed under different but related circumstances, e.g. different points in time. The latter means the graph signals are free from the i.i.d. assumptions required by the classical graph learning models. Experiments on both synthetic and real-world data show that our methods outperform the state-of-the-art models in learning a meaningful graph topology from graph signals, in particular under heavy noise, missing values, and multiple dependency.
Recording sensor data is seldom a perfect process. Failures in power, communication or storage can leave occasional blocks of data missing, affecting not only real-time monitoring but also compromising the quality of near- and off-line data analysis. Several recovery (imputation) algorithms have been proposed to replace missing blocks. Unfortunately, little is known about their relative performance, as existing comparisons are limited to either a small subset of relevant algorithms or to very few datasets or often both. Drawing general conclusions in this case remains a challenge. In this paper, we empirically compare twelve recovery algorithms using a novel benchmark. All but two of the algorithms were re-implemented in a uniform test environment. The benchmark gathers ten different datasets, which collectively represent a broad range of applications. Our benchmark allows us to fairly evaluate the strengths and weaknesses of each approach, and to recommend the best technique on a use-case basis. It also allows us to identify the limitations of the current body of algorithms and suggest future research directions.
Facial action units allow an objective, standardized description of facial micro movements which can be used to describe emotions in human faces. Annotating data for action units is an expensive and time-consuming task, which leads to a scarce data situation. By combining multiple datasets from different studies, the amount of training data for a machine learning algorithm can be increased in order to create robust models for automated, multi-label action unit detection. However, every study annotates different action units, leading to a tremendous amount of missing labels in a combined database. In this work, we examine this challenge and present our approach to create a combined database and an algorithm capable of learning under the presence of missing labels without inferring their values. Our approach shows competitive performance compared to recent competitions in action unit detection.
In data streams, the data distribution of arriving observations at different time points may change - a phenomenon called concept drift. While detecting concept drift is a relatively mature area of study, solutions to the uncertainty introduced by observations with missing values have only been studied in isolation. No one has yet explored whether or how these solutions might impact drift detection performance. We, however, believe that data imputation methods may actually increase uncertainty in the data rather than reducing it. We also conjecture that imputation can introduce bias into the process of estimating distribution changes during drift detection, which can make it more difficult to train a learning model. Our idea is to focus on estimating the distance between observations rather than estimating the missing values, and to define membership functions that allocate observations to histogram bins according to the estimation errors. Our solution comprises a novel masked distance learning (MDL) algorithm to reduce the cumulative errors caused by iteratively estimating each missing value in an observation and a fuzzy-weighted frequency (FWF) method for identifying discrepancies in the data distribution. The concept drift detection algorithm proposed in this paper is a singular and unified algorithm that can handle missing values, but not an imputation algorithm combined with a concept drift detection algorithm. Experiments on both synthetic and real-world data sets demonstrate the advantages of this method and show its robustness in detecting drift in data with missing values. These findings reveal that missing values exert a profound impact on concept drift detection, but using fuzzy set theory to model observations can produce more reliable results than imputation.
Data imputation is the most popular method of dealing with missing values, but in most real life applications, large missing data can occur and it is difficult or impossible to evaluate whether data has been imputed accurately (lack of ground truth). This paper addresses these issues by proposing an effective and simple principal component based method for determining whether individual data features can be accurately imputed - feature imputability. In particular, we establish a strong linear relationship between principal component loadings and feature imputability, even in the presence of extreme missingness and lack of ground truth. This work will have important implications in practical data imputation strategies.